AI video and image news, November 2026: Kling 4.0, GPT Image 2.6, Seedream 5.5 and more
By AI-0 editorial team

On this page
Last updated: November 14, 2026
The biggest AI video news from October 15 to November 14, 2026 was Kling 4.0, with single clips of up to 30 seconds at up to 4K, and Gemini Omni 1.2 Flash, with 15-second clips and a 30% price cut. In images, OpenAI released GPT Image 2.6, ByteDance released Seedream 5.5 and MiniMax opened H3-Image's weights. Only Seedream 5.5 is in AI-0 today.
Every release post this month promised more realism and more consistency, mostly backed by the company's own tests. We use these kinds of models for AI-0's own Instagram and blog, so for each release we say what changed, who it suits, what we think, and plainly whether it's in the app.
What changed in AI video and image models this month
| Model | Company | Date | What's new | Where to use it |
|---|---|---|---|---|
| Kling 4.0 | Kuaishou | Oct 15 | Clips of 3 to 30 seconds at up to 4K, up to 10 keyframes, up to 15 references | Kling (not in AI-0) |
| FLUX 3 Max | fal with Black Forest Labs | Oct 19 | Very fast 720p video, post-trained from FLUX 3 | fal API (not in AI-0) |
| Pika H3 Flex | Pika | Oct 29, app Nov 2 | Post-trained MiniMax H3, 15-second clips with sound, audio stems | Pika (not in AI-0) |
| GPT Image 2.6 | OpenAI | Nov 4 | 50% faster, animated GIFs, SVG export and markup in the API | ChatGPT, OpenAI API (not in AI-0) |
| Gemini Omni 1.2 Flash | Google DeepMind | Nov 5 | 15-second clips, editing over several turns, 30% cheaper | Gemini API, Flow, Gemini app (not in AI-0) |
| Seedream 5.5 | ByteDance Seed | Nov 9 | More natural skin, sharper text, transparent backgrounds, stable edits | AI-0 |
| Firefly update | Adobe | Nov 10 | Creative Agent, Firefly Graph, video editing, Edit Text and Relight, more partner models | Adobe Firefly |
| MiniMax H3-Image | MiniMax | Nov 11 | Open-weight image model, 4K in under 10 seconds, up to 16 references | Hailuo AI, MiniMax API (not in AI-0) |
We didn't find a dated new model release in this window from Veo, Sora, Seedance, Wan, Runway, Grok Imagine, Midjourney or Luma.
What's new in Kling 4.0?
Kuaishou made Kling 4.0 available to everyone on October 15, according to its release notes. It does text-to-video, image-to-video, first and last frames, and up to 10 keyframes in one clip. Clips run from 3 to 30 seconds at 720p, 1080p or 4K, with stereo audio and better lip sync. Omni Reference takes up to 15 inputs at once, including up to 10 images and 5 videos, and you can make targeted edits to an expression or a camera move while the rest of the shot stays put. The 10-bit HDR output is marked "coming soon". Kling 4.0 Flash is the lower-cost version, at 720p and up to 20 seconds.
It suits ads, short dramas and music videos that need one long take. On paper it's the most complete video release of the month, and for social video the 30 seconds matter more than the 4K, because the cut between two generated clips is where a face usually changes. We haven't run Kling 4.0 ourselves, so that's a reading of the spec sheet.
AI-0 has no Kling models. For longer single takes in AI-0, Wan 3.0 Prime goes up to 17 seconds and MiniMax H3 MAX up to 15 seconds, at the time of writing.
What is FLUX 3 Max?
FLUX 3 Max is a video model that fal Research post-trained from Black Forest Labs' FLUX 3, launched on October 19 (fal's announcement). fal says it generates a 10-second clip in about 12 seconds and ties the top image-to-video models in a preliminary Artificial Analysis ranking. It runs at 720p through fal's API, so it's for developers. It isn't in AI-0, where FLUX 3 Klein is a separate image model from the same family.
What is Pika H3 Flex?
Pika announced H3 Flex on October 29 and added it to the Pika app on November 2 (Pika's post). It's MiniMax H3, post-trained to generate in far fewer steps. It drafts clips of up to 15 seconds with synced audio at 480p or 768p, then upscales the one you like to 1080p, 2K or 4K. It takes up to 12 references, and through the API you can get five separate audio stems: dialogue, score, sound effects, Foley and ambience. Pika also released its adapters as open weights. The stems are the part we'd want most: if a clip has good dialogue and bad music, you swap the score in an editor instead of starting again.
Pika H3 Flex isn't in AI-0, but the model it's built on is. MiniMax H3 MAX in AI-0 makes 5 to 15-second clips at 768p or 2K with stereo sound, and takes up to 10 reference images plus audio references at the time of writing.
What's new in GPT Image 2.6?
OpenAI released Images 2.6 on November 4 (OpenAI's announcement). OpenAI says it's 50% faster than before, with two API versions: GPT-Image-2.6 Flare for speed and GPT-Image-2.6 Sunburst for detail and consistency across edits. It can now make animated GIFs. In the API, developers also get markup for targeted edits, SVG export and a complexity setting. It's in ChatGPT and Codex on all subscription tiers.
SVG export is the change designers will feel. GIFs suit reactions and mascots, but a Reel still needs a video model. In our own tests, clothes and jewellery from a reference photo sometimes leaked into the result. OpenAI says 2.6 holds characters better across edits. We haven't tested whether that fixes the leak.
GPT Image 2.6 isn't in AI-0. GPT Image 2.5 Flare and Sunburst are, with up to 8 reference images and transparent backgrounds at 1K at the time of writing.
What's new in Gemini Omni 1.2 Flash?
Google DeepMind rolled out Gemini Omni 1.2 Flash on November 5 (Google's post). Clips now run from 3 to 15 seconds, up from 10 in Omni 1.1. It follows image, motion-clip and audio references more closely, drifts less, and handles dialogue and lip sync better. You can also edit a clip over several turns, changing the background or light without generating it again. In the Gemini API, 720p now costs $0.07 a second, 30% less than before. It's generally available in the Gemini API and Google Cloud and rolling out in Flow, Google Vids and the Gemini app, where 15-second videos go first to Google AI Ultra subscribers.
Fixing one detail without re-rolling the whole clip saves more money than five extra seconds, so the editing is the real upgrade. We also like that Google admits fast sports action and complex physics still need work.
Gemini Omni isn't offered in AI-0 right now. For reference-driven clips in the app, Wan 3.0 Prime takes image, video and audio references at the time of writing.
What's new in Seedream 5.5?
ByteDance Seed released Seedream 5.5 on November 9 (ByteDance's post). Light, materials and skin look more natural and less glossy, and small text renders more cleanly. Across repeated edits the untouched parts stay stable, for 100 rounds in ByteDance's demo. It can also output real transparent backgrounds. ByteDance's own tests put it ahead of GPT Image 2.5 Sunburst and Nano Banana 2.1 on text and realism.
This is the release we've used most. We made our Instagram carousel covers and our blog covers with it this month. One lesson: to leave room for a title, we had to ask for "the top 45% empty plain wall". Asking for the subject "further back" wasn't enough.
You can use it in AI-0 now. Tap Image, tap the model row and choose Seedream 5.5. At the time of writing it offers 1K, 1.5K or 2K and a Background option you can set to Transparent. The app always shows the current options.
What did Adobe announce for Firefly?
Adobe expanded Firefly at Adobe MAX on November 10 (Adobe's announcement). There's no new Adobe model here. The Creative Agent now works across Firefly with a plan you approve first, Firefly Graph (beta) saves repeatable workflows, the video editor gained a timeline and captions, and the image editor added Edit Text, Relight and Retouch. Firefly now also offers other companies' models, including Kling 4.0, GPT Image 2.6 and Nano Banana 2.1, so Adobe is becoming a front end for the labs it competes with. It suits desktop teams on Creative Cloud. Firefly isn't part of AI-0.
What is MiniMax H3-Image?
MiniMax released H3-Image on November 11 (MiniMax's post), an image model built on H3 with open weights. MiniMax says it makes a 4096 by 4096 image in under 10 seconds and understands up to 16 reference images. It's on Hailuo AI and the MiniMax API. The open weights are the news, because anyone with the hardware can run and customise it. It isn't in AI-0, though MiniMax's video model H3 MAX is.
What this means if you make videos on your phone
Clips are getting longer in one generation, and models are learning to edit what they made instead of starting over. Both mean fewer retries, which is what eats credits.
Most new features sit behind APIs or paid tiers, and the prompt matters more than the version. Every clip prompt we write names the shot size, lens, camera rig and move, speed and end framing, and we test one clip before running a batch, because a bad batch wastes credits.
Our picks in AI-0 today:
| Goal | Model in AI-0 |
|---|---|
| A photoreal still to animate, or a cover with room for text | Seedream 5.5 or GPT Image 2.5 |
| A video that follows several reference images, clips or audio | Wan 3.0 Prime |
| A 2K clip with stereo sound or detail | MiniMax H3 MAX |
| A longer single take, up to 17 seconds | Wan 3.0 Prime |
| A 5-second 720p clip from one image | Runway Gen-4 Turbo |
| A 6 or 10-second clip from an image, with sound added for you | Grok Imagine |
Video costs far more credits than images, and longer or higher-resolution clips cost more, so animate only the still you're sure about. Our guide to text-to-video vs image-to-video explains when each works better, and how AI-0 credits work covers the costs.
Frequently asked questions
What is the best AI video model right now?
No single model wins at everything. Of this month's releases, Kling 4.0 has the longest clips and most reference options on paper, and Gemini Omni 1.2 Flash is cheaper to edit with. In AI-0 today, we'd pick Wan 3.0 Prime for reference-driven shots and MiniMax H3 MAX for 2K clips with sound.
Is Kling 4.0 free?
Kling's release notes for 4.0 don't describe a free allowance, so check Kling's own pricing before you count on one. Kling 4.0 Flash is the lower-cost version, limited to 720p and 20 seconds. Kling isn't available in AI-0.
What's new in GPT Image 2.6?
OpenAI says GPT Image 2.6 is 50% faster, keeps characters and products more consistent across edits, and can make animated GIFs. In the API it adds markup for targeted edits, SVG export and a complexity setting. AI-0 has GPT Image 2.5 Flare and Sunburst, not 2.6.
Is Gemini Omni 1.2 Flash better than the earlier Gemini Omni?
Google says yes: clips go up to 15 seconds instead of 10, drift less and can be edited over several turns, at 30% less in the API than Omni 1.1. Gemini Omni isn't offered in AI-0 right now.
Can I use Seedream 5.5 for free?
Yes. In AI-0 you can use Seedream 5.5 with credits earned by watching optional ads. New accounts start with 15 credits and each ad adds 5, and the cost of each image is shown on the Generate button.
Try the one you can use today
Open AI-0, make one still with Seedream 5.5, then turn it into a short clip with Wan 3.0 Prime or MiniMax H3 MAX. AI-0 is free to download on the App Store and Google Play, and the free credits guide shows how to earn more as you go.
Create on your phone with AI-0
Watch a short ad when you need credits, choose a video or image model, and create without a monthly subscription.


